Visible light positioning (VLP) has been widely studied since it can provide high accuracy indoor positioning functionality with light-emitting diode (LED) beacons. However, existing VLP systems still suffer from high positioning latency and low robustness in practical scenarios. A robust and low-complexity image processing algorithm is proposed for LED region-of-interest (ROI) extraction and tracking, which contributes to handle various harsh situations and support accurate VLP with low positioning latency. Experimental results show that the proposed LED-ROI scheme can achieve an average tracking accuracy of 0.46 cm and better real-time performance and robustness compare with the others current LED-ROI methods.